Inter-rater agreement of study characteristics and quality ratings was assessed using descriptive statistics and intra-class correlation coefficient (ICC). A random effects model using inverse variance weights examined the ES of CBT and SRIs in CMA.[66] A random effects model was chosen because the true ES were expected to vary across trials due to different study characteristics.[68] Heterogeneity of ES was assessed using the forest plot, Q statistic, and I2 statistic. Publication bias was assessed by visual inspection of the funnel plot and Egger’s test for bias. When publication bias was present, Duval and Tweedie’s trim-and-fill method was used to account for publication bias by producing an adjusted summary effect that takes into account potential within the field.[68] An analog to the analysis of variance (ANOVA) examined the heterogeneity of ES across comparison conditions (non-active versus active comparison conditions). Separate random effect models examined the RR of CBT and SRI in CMA for treatment response and symptom/diagnostic remission. The same procedures noted above assessed for publication bias and sensitivity analyses. The number needed to treat (NNT) was calculated for treatment response
